Static random access memories (SRAMS) are one type of known volatile semiconductor memory device. The SRAM typically has memory cells formed at points of intersection between complementary data lines (bit lines) and word lines arranged in a matrix fashion. FIG. 32 is an equivalent circuit diagram showing a memory cell of a conventional SRAM. A typical circuit constitution of the SRAM will now be described with reference to FIG. 32.
A memory cell of the conventional SRAM comprises two access transistors A1 and A2, two driver transistors D1 and D2, and two load transistors P1 and P2. The two load transistors P1 and P2 and the two driver transistors D1 and D2 make up a flip-flop circuit. The flip-flop circuit constitutes two memory nodes N1 and N2 that are cross-coupled. The memory nodes N1 and N2 are in a bi-stable state involving one of two settings: either N1 is High and N2 Low, or N1 is Low and N2 High. The bi-stable state is maintained as long as power is supplied appropriately.
One source/drain region of each of the access transistors A1 and A2 is connected to the memory nodes N1 and N2 respectively, i.e., to input/output terminals of the flip-flop circuit. The other source/drain regions of the access transistors A1 and A2 are connected to bit lines. The gate electrodes of the access transistors A1 and A2 are connected to a word line. The word line controls the activation and deactivation of the access transistors A1 and A2.
The drain regions of the driver transistors D1 and D2 are connected to the source/drain regions on one side of the access transistors A1 and A2 respectively. The source regions of the driver transistors D1 and D2 are connected to ground (VEE line). The gate electrode of the driver transistor D1 is connected to the source/drain region of the access transistor A2, and the gate electrode of the driver transistor D2 is coupled to the source/drain region of the access transistor A1. The source/drain regions on one side of the load transistors P1 and P2 are connected to the source/drain regions on one side of the access transistors A1 and A2. The source/drain regions on the other side of the load transistors P1 and P2 are connected to power supply line (VCC line).
For a data write operation, the word line (WL) is selected so as to turn on the access transistors A1 and A2. A voltage is then forced onto the bit line pair in accordance with a desired logical value, whereby the flip-flop circuit has its bi-stable state brought into one of the two settings described above.
For a data read operation, the access transistors A1 and A2 are first turned on. Then a potential on the memory nodes N1 and N2 is transmitted to the bit lines.
Having the above-described constitution, so-called six-transistor type SRAM cells that use PMOSs as load transistors formed on a substrate are extensively practiced today (called full-CMOS type SRAM cells hereunder). In a full-CMOS type SRAM cell, the drain region (P.sup.+ diffusion region) of the PMOSs (load transistors) constituting one inverter of the flip-flop circuit needs to be connected to the drain region (N.sup.+ diffusion region) of the NMOSs (driver transistors).
Traditionally, full-CMOS type SRAM cells utilize metal lead providing ohmic contacts for interconnection, i.e., local lead that encompasses all lead arrangements connecting diffusion layers of transistors as well as contiguous elements. Illustratively, Japanese Patent Laid-Open No. Hei 9-55440 discloses a full-CMOS type SRAM that uses metal arrangements for lead. The disclosed semiconductor device has tungsten-embedded electrodes connecting metal lead layers to the substrate, and also has tungsten-embedded electrodes connecting contact holes that link contiguous elements.
Generally, metal lead has poor workability. Pattern pitches can be reduced to only a certain extent, so that it is difficult to achieve further miniaturization. Another disadvantage of metal lead-in general is its poor resistance to heat. This limits the scope of thermal treatment following pattern formation.
One proposed solution to the above difficulties with conventional full-CMOS type SRAM cells involves the use of a polycrystal silicon film as lead that connects specifically the drain regions of PMOSs (load transistors) to those of NMOSs (driver transistors). However, the use of such a polycrystal silicon film in the conventional SRAM poses other problems, as will be described below.
FIGS. 33 and 34 are an equivalent circuit diagram and a cross-sectional view of the conventional SRAM respectively, both intended to explain problems therewith. In FIG. 34, reference numeral 51 is an N.sup.- -type silicon substrate, 52 is a P-type well region, 53 is an N-type well region, and 54 is a field insulating film for isolating elements. Driver transistors are formed on the surface of the P-type well region 52 surrounded by the field insulating film 54. Each driver transistor is composed of N.sup.+ -type source/drain regions 55a and 55b, N.sup.- -type source/drain regions 56a through 56c, a gate oxide film 58, a gate electrode 59a, and a side wall oxide film 60.
Load transistors are formed on the surface of the N-type well region 53 surrounded by the field insulating film 54. Each load transistor is made up of a P.sup.+ -type source/drain region 57, a gate oxide film 58, a gate electrode 59b, and a side wall oxide film 60. The entire surface is covered with a silicon dioxide film 61. Contact holes 62a and 62b are formed on the N.sup.+ -type source/drain region 55b of the driver transistor and on the P.sup.+ -type source/drain region 57 of the load transistor. A polycrystal silicon film 63 is formed inside the contact holes 62a and 62b as well as over the silicon dioxide film 61. The polycrystal silicon film 63 is a P-type polycrystal silicon film doped with P-type impurities such as boron. The P-type polycrystal silicon film connects the N.sup.+ -type source/drain region 55b of the driver transistor to the P.sup.+ -type source/drain region 57 of the load transistor.
It should be noted that transistor-to-transistor connection by means of single-layer lead leads to connecting the drain regions of PMOSs (load transistors) to those of NMOSs (driver transistors) through a conductive polycrystal silicon film. In the interconnection utilizing a single-layer polycrystal silicon film of the same conductivity, as shown in FIGS. 33 and 34, subsequent heat treatment causes the polycrystal silicon film to diffuse impurities into the silicon substrate. This leads to another problem: PN diodes are formed in the silicon substrate.
The PN diodes are produced because P-type impurities contained in the polycrystal silicon film 63 are diffused into the substrate, thereby generating a P.sup.+ -type diffusion region 64 in the N.sup.+ -type source/drain region 55b. This allows the High level of the memory nodes N1 and N2 to rise just up to VCC-Vbi (Vbi: built-in potential of PN junction.apprxeq.0.8 V), which tends to make the High level of the memory nodes unstable. With the High node level in such an unstable state, a drop in the resistance to so-called soft errors may become pronounced.
Soft errors can occur as follows: alpha rays upon entry from the outside such as the package material produce electron-hole pairs. Of these pairs, electrons are attracted to memory nodes in the memory cells and invert stored data in the cells, causing random errors called soft errors. When the High node level of the memory cells drops dragged by a falling potential accumulated therein, the resistance to such soft errors will deteriorate.
When the polycrystal silicon film connects the drain regions of PMOSs (load transistors) to those of NMOSs (driver transistors), the presence of PN diodes that may be formed in the silicon substrate and a high lead resistance of the silicon film itself combine to boost the resistance of connection between the drain regions of the PMOSs (load transistors) and those of the NMOSs (driver transistors). The enhanced connection resistance makes it difficult to supply charges to the memory nodes, reducing the accumulated charges in the High level nodes of the memory cells. This further promotes the possibility of causing soft errors.
Full-CMOS type SRAM cells are subject to another disadvantage: each cell must comprise two PMOSs and four NMOSs. That transistor layout requires the full-CMOS type SRAM cell to have a greater cell area than other types of SRAMs.